IoT and Context-Aware Mobile Recommendations Using Multi-agent Systems

Abstract

The work presents a Multi-Agent System (MAS) solution for generating personalized recommendations on mobile devices with the use of contextual data acquired from the Internet of Thing (IoT). For mass data processing and personalized model preparation a server-side MAS architecture for Big Data processing is used. In order to deliver context-aware recommendations for very sparse input data, based on different sensors and IoT information, the usage of CARS2, state-of-the-art factorization technique is proposed. The main assumption regarding mobile device in the presented architecture is unreliable communication channel between mobile and server side. This is the result of the novel paradigm of computing model-base machine learning computations called "Model to Data", where in case of the recommender systems all the computations for active user and current context is handled by mobile device themselves only using previously acquired necessary precomputed parts of the model from the server side. As a example of a real-life application the dietary/fitness recommender system is included.